Table 42

ai-sanctions-wave-part4

Posted on by root

AI Sanctions Wave – Part 4: The Judicial AI Paradox

Courts are sanctioning lawyers for AI-generated fake citations. Meanwhile, 61.6% of federal judges report using AI themselves—for the same functions.

—

In March 2026, as courts imposed $145,000 in sanctions on lawyers for filing AI-generated fake citations, a Northwestern University study published in the Sedona Conference Journal revealed a striking counterpoint: 61.6 percent of federal judges report using at least one AI tool in their judicial work.[1]

These judges use AI primarily for legal research (30 percent of users) and document review (15.5 percent)[2]—the same categories of work that, when performed poorly by attorneys, trigger the sanctions described in earlier parts of this series.

The asymmetry is undeniable. Lawyers face monetary penalties, disciplinary referrals, and professional consequences for inadequate verification of AI-generated content. Judges report substantial AI adoption for similar workflows but face no verification standards, no disclosure requirements, and no ethical obligations comparable to lawyers’ duty of candor to the tribunal.

This installment examines the judicial AI paradox: why lawyers are sanctioned for AI-assisted errors while judges use AI for the same functions without parallel obligations, what this asymmetry means for the legal system, and whether courts should address whether judicial verification obligations exist.

The question is not whether judges should use AI. That ship has sailed. The question is whether courts will impose verification standards on judicial AI use that match what they require of counsel.

—

The Northwestern Study: Baseline Data

The Northwestern University study, authored by Daniel Linna of Northwestern Pritzker Law and V.S. Subrahmanian of the Northwestern Security & AI Lab, represents the first random-sample survey of federal judges on AI use.[3]

Methodology and Sample: – Stratified random sample of 502 federal judges – Response rate: 112 responses (22.3%) – Survey conducted: December 2-19, 2025 – Publication: Sedona Conference Journal, March 2026

The study’s findings establish a baseline for understanding judicial AI adoption:

Adoption Rates: – 61.6%: Use at least one AI tool in judicial work – 22.4%: Use daily or weekly (habitual users) – Remainder: Occasional users
Use Cases: – 30% of users: Legal research (the same function lawyers use AI for before submitting briefs) – 15.5% of users: Document review (evaluating motions, briefs, and discovery materials) – Other uses: Summarization, drafting, administrative tasks
Training Gap: – 45.5% of judges reported that court administration provided no AI training – Most judges described themselves as self-taught or learning informally
Attitudinal Split: – Roughly half expressed optimism about AI’s potential – Roughly half expressed concern about AI risks – Views were evenly divided with no clear consensus

The study is methodologically significant for three reasons. First, the random-sample design means the findings are nationally representative across court levels, regions, and demographics. Second, peer-reviewed publication in the Sedona Conference Journal adds academic credibility to the data. Third, the survey establishes a baseline for longitudinal studies tracking how judicial AI use evolves over time.[4]

What the study does not do is address the question that follows logically from its findings: If 61.6% of judges use AI for legal research and document review, what verification obligations, if any, attach to that use?

—

The Asymmetry: Lawyers Sanctioned, Judges Unconstrained

The asymmetry is stark:

What Lawyers Face: – $145,000+ in Q1 2026 sanctions for AI-generated fake citations[5] – Mandatory verification standards enforced through sanctions – Professional responsibility rules (ABA Formal Opinion 512, Model Rules 1.1, 3.3, 5.1, 5.3)[6] – Disciplinary consequences: Fines, adverse costs, bar referrals, disqualification, public admonishment[7] – Citation accuracy standards: Oregon’s $500/citation, $1,000/quotation formula; Sixth Circuit’s tool-agnostic principle[8]
What Judges Face: – No judicial verification standards established for AI-generated content – No disclosure requirements for judicial AI use in opinions or orders – No ethical obligation under current canons of judicial conduct similar to lawyers’ duty of candor to tribunals[9] – No disciplinary mechanism for judicial AI errors comparable to lawyer sanctions – Judicial immunity protects judicial acts from liability, extending to AI-assisted opinions[10]

The question raises itself: Why are lawyers sanctioned for AI-assisted hallucinations while judges use AI for the same functions—legal research, document review, citation generation—without verification standards or enforcement consequences?

Four explanations are commonly offered for this asymmetry. Each has merit. Each also has counterarguments that complicate the picture.

—

Explanation 1: Different Roles, Different Duties

The Argument: Lawyers and judges perform fundamentally different roles in the legal system with asymmetric duties justifying asymmetric enforcement.

Lawyers are advocates who owe a duty of candor to the tribunal, meaning they must not mislead the court through false statements or fabricated evidence.[11] When a lawyer files a brief containing fake citations, that duty is violated directly. The lawyer is presenting the court with misinformation.

Judges, by contrast, are decision-makers who owe a duty to apply the law correctly to the facts before them.[12] A judge’s AI-assisted opinion is not an advocacy statement submitted to a tribunal—it is the tribunal’s own reasoning. The duty framework is different.

The Counterargument: The different-roles argument assumes that judicial AI errors are less consequential than lawyer errors, but this may be mistaken.

Consider the precedential impact. When a lawyer files a brief with fake citations, the error is contained to that case and that litigant. The court identifies the error and sanctions the lawyer. The damage is localized.

When a judge issues an opinion relying on AI-generated fake citations, that opinion becomes precedent. Future lawyers cite that precedent. Other judges rely on it. The error propagates through the legal system. The potential systemic impact of judicial AI errors may be higher than lawyer errors, not lower.

Moreover, the reliance interest is different. Lawyers submit briefs expecting judicial review—judges are expected to verify citations and catch errors. But who verifies the judge? Appellate review exists, but not all cases are appealed. Unpublished opinions receive limited scrutiny. Trial court orders often become final without review.

If reliance on verification is asymmetric (judges verify lawyers; no one verifies judges), the different-roles argument becomes circular: we enforce asymmetric standards because roles are different; roles are different because we enforce asymmetric standards.

—

Explanation 2: Self-Correction Mechanisms

The Argument: The legal system has built-in self-correction mechanisms that address judicial AI errors differently than lawyer errors.

Lawyers’ errors are subject to immediate judicial review—every brief, motion, and filing is examined by a judge who can identify fake citations and sanction the lawyer. Sanctions serve as a substitute for direct correction when judges do catch errors.

Judges’ errors are corrected through appellate review. If a trial judge makes an error—whether human or AI-assisted—the affected party can appeal. The appellate court reviews the decision, identifies errors, and reverses or remands. In this view, appellate availability substitutes for direct enforcement against judges.

The Counterargument: Self-correction is not a reliable enforcement mechanism for three reasons.

First, not all lawyer errors are detected. Courts have limited resources for citation verification—a judge reading hundreds of motions per month cannot independently verify every citation in every filing. Some AI-generated fake citations undoubtedly slip through undetected, and sanctions only address failures that come to light. Sanctions are retrospective; they compensate for detection gaps only imperfectly.

Second, judicial errors may not be corrected through appellate review. Unpublished opinions cannot be cited as precedent, limiting their systemic impact—but they also receive less appellate scrutiny. Trial court orders that settle cases become final without appeal. Procedural bars (waiver, forfeiture, timeliness) prevent some errors from being raised on appeal.

Third, appellate review is reactive, not preventive. The Sixth Circuit’s $30,000 sanction in Whiting v. City of Athens was preventive—expressly intended to deter future misconduct by raising the cost of non-verification.[13] If appellate review is the only correction mechanism for judicial AI errors, the system is purely reactive. Lawyer sanctions prevent future errors; judicial self-correction reacts to past errors after damage is done.

The self-correction argument assumes that appellate review is equivalent to direct judicial oversight of lawyer filings, but the mechanisms are fundamentally different. One is prospective enforcement; the other is reactive correction.

—

Explanation 3: Educational vs. Enforcement Phase

The Argument: Lawyers are in the enforcement phase of AI adoption while judges remain in the educational phase, justifying the asymmetry in standards.

Lawyers have had three years of warnings and education since Mata v. Avianca (2023) made AI hallucinations a national story.[14] Courts initially took an educational approach—warnings, orders to show cause, judicial opinions explaining the problem—but found that warnings proved insufficient. Only after that educational period did courts escalate to monetary sanctions. Q1 2026’s $145,000 in penalties represents enforcement phase kicking in.

Judges, by contrast, are newer to AI use. The Northwestern study found that 45.5% of judges received no formal AI training, and most are self-taught or learning informally.[15] In this view, courts remain in an early learning phase for judicial AI use—a period of experimentation, data gathering, and understanding capabilities before establishing standards. Imposing enforcement now would be premature.

The Counterargument: The educational-enforcement phase argument assumes that lawyers and judges followed parallel AI adoption timelines, but the evidence suggests judges had a longer observation window.

Lawyers’ three-year learning curve began in 2023 with Mata v. Avianca, but judges observed those cases firsthand. Federal district judges saw the early AI hallucination cases, sanctioned lawyers, and developed responses. State court judges did the same. By the time lawyers entered the enforcement phase in 2026, judges had been watching AI-related misconduct for three years.

Moreover, professional responsibility rules—ABA Formal Opinion 512, Model Rules 1.1 and 3.3—apply to lawyers and establish clear standards for AI use.[16] Parallel judicial conduct standards do not exist. The asymmetry is not temporal (learning vs. enforcement) but structural (standards exist for lawyers, not judges).

Even accepting the phase-difference argument, the question remains: Why are judicial AI verification standards not being developed now in anticipation of an eventual enforcement phase? Waiting until judges make AI-generated errors before establishing standards guarantees reactive correction rather than preventive enforcement.

—

Explanation 4: Political Feasibility

The Article: Disciplining judges is politically and institutionally difficult in ways that sanctioning lawyers is not.

Judges enjoy lifetime tenure (Article III federal judges) and substantial independence protections designed to preserve judicial decision-making free from political pressure.[17] Judicial conduct commissions, which investigate complaints against judges, are slow, cautious, and rarely impose meaningful sanctions—particularly for errors in judicial reasoning rather than corruption or misconduct.

Lawyers, by contrast, face readily available enforcement mechanisms. Courts can sanction lawyers immediately without involving disciplinary bodies. State bar associations have established procedures for addressing misconduct. Sanctions are part of everyday judicial administration.

From this perspective, the asymmetry reflects institutional reality: it is feasible to sanction lawyers for AI errors; it is not currently feasible to hold judges to similar standards, even if desired.

The Counterargument: Political feasibility addresses enforcement difficulty, not ethical justification. The asymmetry remains a fairness question even if correcting it is politically difficult.

The judicial AI paradox creates a credibility problem. Courts impose verification standards on lawyers while judges themselves use AI without parallel obligations. This double standard undermines the legitimacy of enforcement—why should lawyers take verification duties seriously when the judges enforcing them operate under different rules?

Even if direct disciplinary enforcement against judges is politically infeasible, courts have other options. Judicial conduct commissions could issue advisory opinions addressing AI verification obligations. The Federal Judicial Center could develop training programs emphasizing verification parallel to lawyer standards. Model Codes of Judicial Conduct could be amended to include verification duties.

Political feasibility explains why the asymmetry persists but does not justify it. The fairness question remains unresolved.

—

Practical Implications: Where the Asymmetry Matters

The judicial AI paradox is not merely a theoretical fairness concern. It has practical implications for how litigation is conducted and how outcomes are determined.

eDiscovery and Technology-Assisted Review (TAR)

Technology-Assisted Review (TAR) workflows rely on AI algorithms to identify responsive documents in discovery. Parties submit TAR outputs as evidence, and courts evaluate TAR methodology under judicial supervision.

Current practice reveals the asymmetry:

– Lawyer-submitted TAR: Scrutinized for methodology, transparency, and verification. Courts can sanction lawyers for inadequate TAR validation or for failing to disclose AI use in document production.[18] – Judicial evaluation of TAR: No verification obligation when judges use AI to evaluate discovery disputes or analyze document productions. If a judge’s AI mischaracterizes a TAR methodology, there is no enforcement mechanism.

For eDiscovery professionals, this creates uncertainty. If parties invest resources in TAR validation to avoid sanctions, but judges use AI to evaluate that work without verification standards, the investment may be offset by inconsistent judicial oversight.

Motion Practice

Motion practice shows the asymmetry in its clearest form:

– Lawyer motions using AI: Must verify every citation. A single fake citation can trigger $500-$1,000 in sanctions under Oregon’s formula, or steeper penalties in federal court.[19] – Judicial rulings using AI: No verification standard. Citation errors in judicial opinions are corrected on appeal, but not prevented through direct enforcement.

This asymmetry affects strategic calculus. Lawyers filing motions must invest in verification both to avoid sanctions and to anticipate judicial scrutiny. But if judges use AI to analyze motions without verification obligations, the effort-reward balance shifts. Lawyers may file more motions expecting that AI-assisted judicial scrutiny will be less rigorous than manual review.

Appeals

Appellate practice illustrates the stakes:

– Appellate briefs with AI: Must be verified under heightened appellate standards. Sanctions are steeper at appellate levels—Whiting v. City of Athens imposed $30,000 in federal appellate penalties, and the Sixth Circuit reinforced this line three weeks later in United States v. Farris (April 2026), where CJA counsel was removed and denied compensation for AI-generated misrepresentations.[20] The Fifth Circuit entered the enforcement picture with Fletcher v. Experian Info. Sols., Inc., 168 F.4th 231 (5th Cir. 2026), imposing $2,500 in sanctions for 16 fabricated quotes, with the court noting that dishonesty about AI use triggers harsher penalties. Two federal appellate circuits have now published AI sanctions opinions; others will likely follow. – Appellate opinions may use AI: No verification standard. Errors in appellate opinions become binding precedent, propagating errors through case law.

For litigants considering appeals, this creates added uncertainty. If appellate opinions may contain AI-generated errors that become binding precedent, the cost of appealing increases. Parties must weigh the risk of flawed precedent against the benefits of correcting trial court errors.

Settlement Negotiations

The judicial AI paradox may also affect settlement decisions:

– Parties may be more likely to settle if they suspect judicial AI use introduces uncertainty about outcomes – Risk-averse litigants may prefer certainty of settlement over uncertain appellate review of potential AI-influenced opinions – The asymmetry may increase settlement pressure, not through merits assessment but through uncertainty about judicial fact-finding

—

Should Courts Address This Gap?

Arguments can be made for and against action. Both sides have merit.

Arguments For Addressing the Gap

Fairness: The current asymmetry undermines judicial credibility. Courts imposing verification standards on lawyers while judges operate without parallel obligations creates a double standard that erodes confidence in the fairness of enforcement. Equity demands parallel obligations, even if enforcement mechanisms differ.
Risk Mitigation: Judicial AI errors have higher systemic risk than lawyer errors. Lawyer errors are contained to specific cases and caught through judicial review. Judge errors become precedent, affecting future cases and litigants for years. Early correction—through verification standards—prevents accumulation of problematic precedent that would require costly appellate correction later.
Transparency: Disclosure allows parties to assess the reliability of judicial reasoning. If judges disclosed AI use in opinions, parties could evaluate whether verification was adequate, raise concerns on appeal, or request clarification. Transparency without enforcement still improves appellate review by creating a record.
Proactive Standard-Setting: Addressing the gap proactively allows courts to shape standards rather than reacting to errors. ABA Formal Opinion 512 established lawyer-side standards before sanctions accelerated.[21] Judicial standards could be established similarly, creating a framework for appropriate AI use before problems emerge.

Arguments Against Addressing the Gap

Judicial Independence: Imposing verification requirements could be seen as interfering with judicial discretion. Judges need autonomy in how they conduct research and analysis. Mandating verification procedures might stifle judicial adoption of technology that improves efficiency. Executive branch influence on judicial work methods raises constitutional separation-of-powers concerns.
Administrative Burden: Training the majority of federal judges on AI verification would be costly and time-consuming. Verification procedures could slow judicial decision-making, particularly in high-volume trial courts. Courts lack resources to implement new oversight systems on top of existing caseload pressures.
Different Functions: Judges are triers of fact and law, not advocates. Their errors are corrected through appellate review in ways lawyer errors are not. Judicial errors do not violate duties owed to tribunals because judges are the tribunal. Disciplinary mechanisms exist for judicial misconduct, even if rarely used—judicial immunity does not preclude impeachment or judicial conduct commission action.

Potential Solutions

If courts do address the gap, several approaches are available:

Code of Conduct Amendment: The Model Code of Judicial Conduct could be amended to include a verification obligation for AI use, similar to ABA Formal Opinion 512’s application of Model Rules 1.1 and 3.3 to lawyers.[22] States would adopt the amended canon, and judicial ethics commissions would enforce it through admonishments or censure when violations occur.
Judicial Training Programs: The Federal Judicial Center and state judicial education programs could develop mandatory training on AI verification. Training would emphasize verification duties parallel to those imposed on lawyers—reading every citation, confirming case law through primary sources, disclosing AI use when material to reasoning.
Disclosure Rules Without Verification: Courts could require disclosure of AI use in opinions and orders without imposing verification obligations. Transparency would allow parties to identify AI-assisted reasoning, raise concerns on appeal, and create a record for appellate review. This approach addresses fairness concerns without imposing administrative burdens or touching judicial independence in direct verification enforcement.
Appellate Review Clarification: Courts of appeal could explicitly address whether parties may raise allegations of AI-generated errors as grounds for reversal. Clarifying that appellate courts will inquire about judicial AI use when errors are suspected would create a reactive mechanism for correction without imposing proactive verification duties on judges.

—

The Core Question: Does Judicial Immunity Extend to AI-Assisted Errors?

The unresolved tension underlying the judicial AI paradox is whether judicial immunity covers AI-assisted errors.

Current State: – Judicial immunity is absolute for judicial acts performed within the judge’s jurisdiction, even if erroneous or malicious.[23] – AI-assisted research, document review, and opinion drafting are judicial acts within judicial discretion – No mechanism exists for judicial AI-specific enforcement beyond existing judicial conduct processes (which are rare and slow)
Proposed State: – Verification obligations could be created through Model Code of Judicial Conduct amendments – Enforcement would operate through reputational mechanisms (public criticism, appellate correction) rather than monetary sanctions – Disclosure requirements would allow parties to identify AI use and raise concerns on appeal – Judicial immunity would not be breached—no liability, no fines—but professional obligations would exist
The Tension: Can courts require lawyers to verify AI-generated content under threat of sanctions while judges themselves use AI without verification standards? The Sixth Circuit’s tool-agnostic principle—”no filing should contain citations…that a lawyer has not personally read and verified, regardless of source”[24]—applies explicitly to lawyers. If verification obligations are source-agnostic for lawyers, why not for judges?

The question is not whether judges should use AI—they already do, at scale and in growing numbers.[25] The question is whether professional responsibility will be symmetrical or asymmetrical. The current asymmetry is historically ordinary—lawyers and judges have always faced different standards—but it becomes legally and ethically fraught when both use AI for the same functions yet face wildly different consequences for errors.

—

Sources

—

Continue to Part 5: The Labeling Problem
Back to Part 3: The Sixth Circuit Line
Series Index: AI Sanctions Wave

Notes

  1. Northwestern University, “Federal Judges Report Broad Adoption of AI Tools,” *Northwestern Now*, March 30, 2026, https://news.northwestern.edu/stories/2026/03/northwestern-study-finds-a-significant-number-of-federal-judges-are-already-using-ai-tools ↩

  2. Northwestern University, note 1 (study methodology: random-sample survey; 22.4% use AI weekly/daily; 61.6% use at least one AI tool; legal research 30%, document review 15.5%) ↩

  3. Daniel Linna Jr. and V.S. Subrahmanian, “Federal Judges and AI: Adoption, Attitudes, and Implications,” *Sedona Conference Journal*, Vol. 27, No. 1 (March 2026), https://sedonaconference.org/publications/Sedona-Conference-Journal-Vol-27-No-1 ↩

  4. Northwestern University, note 1 (peer-reviewed random-sample design) ↩

  5. EDRM/ComplexDiscovery, “The AI Sanction Wave: $145K in Q1 Penalties Signals Courts Have Lost Patience with GenAI Filing Failures,” April 6, 2026, https://complexdiscovery.com/the-ai-sanction-wave-145k-in-q1-penalties-signals-courts-have-lost-patience-with-genai-filing-failures/ ↩

  6. American Bar Association Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools and the Profession,” July 29, 2024, https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/aba-formal-opinion-512/ ↩

  7. EDRM/ComplexDiscovery, note 5 (sanctions types and enforcement mechanisms overview) ↩

  8. *Ringo v. Colquhoun Design Studio, LLC*, 345 Or. App. 301 (December 2025), https://law.justia.com/cases/oregon/court-of-appeals/2025/a186670.html ($500/citation, $1,000/quotation); *Whiting v. City of Athens*, No. 25-5424 (6th Cir. March 13, 2026), https://law.justia.com/cases/federal/appellate-courts/ca6/25-5424/25-5424-2026-03-13.html (tool-agnostic principle) ↩

  9. Model Code of Judicial Conduct, 2020 Edition, https://www.americanbar.org/groups/professional_responsibility/publications/model_code_of_judicial_conduct/ (current canons do not address AI verification obligations) ↩

  10. *Stump v. Sparkman*, 435 U.S. 349 (1978), https://supreme.justia.com/cases/federal/us/435/349/ (judicial immunity for judicial acts even when erroneous; AI-assisted reasoning falls within judicial acts) ↩

  11. Model Rules of Professional Conduct, Rule 3.3 (Candor Toward Tribunal), https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_3_3_candor_toward_tribunal/ ↩

  12. Model Code of Judicial Conduct, Canon 2, note 9 (“A judge shall uphold and promote the independence, integrity, and impartiality of the judiciary”) ↩

  13. *Whiting v. City of Athens*, note 8, at *8 (court cited prior inadequate sanctions as justification for elevated $30,000 penalty) ↩

  14. *Mata v. Avianca, Inc.*, No. 22-cv-1461 (PKC) (S.D.N.Y. June 22, 2023), https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/ ↩

  15. Northwestern University, note 1 (45.5% of judges received no AI training) ↩

  16. ABA Formal Opinion 512, note 6 (applies Rules 1.1, 3.3, and supervisory obligations to AI use) ↩

  17. 28 U.S.C. § 371 (Article III judicial tenure protections), https://www.law.cornell.edu/uscode/text/28/371 ↩

  18. EDRM/ComplexDiscovery, note 5 (court scrutiny of TAR workflows) ↩

  19. *Ringo v. Colquhoun Design Studio, LLC*, note 8; see also Part 2 of this series for detailed analysis ↩

  20. *Whiting v. City of Athens*, note 8 (federal appellate sanctions precedent); *United States v. Farris*, 6th Cir. (April 3, 2026) (CJA counsel removed, denied compensation for AI-generated misrepresentations); *Fletcher v. Experian Info. Sols., Inc.*, 168 F.4th 231 (5th Cir. 2026) ($2,500 sanctions for 16 fabricated quotes; dishonesty about AI use triggers harsher penalties) ↩

  21. ABA Formal Opinion 512, note 6 (proactive standard-setting for lawyer AI use) ↩

  22. Model Code of Judicial Conduct, note 9 (framework for potential amendment on AI verification) ↩

  23. *Stump v. Sparkman*, note 10 (absolute judicial immunity doctrine) ↩

  24. *Whiting v. City of Athens*, note 8 ↩

  25. Northwestern University, note 1 (61.6% adoption rate among responding federal judges) ↩

  26. Reuters, “Majority of US federal judges are using AI, study finds,” March 30, 2026, https://www.reuters.com/legal/government/majority-us-federal-judges-are-using-ai-study-finds-2026-03-30/ ↩

  27. Law.com LegalTechNews, “Some Federal Judges Are Embracing Gen AI, Though Many Are Split on Its Potential for Courts,” March 30, 2026, https://www.law.com/legaltechnews/2026/03/30/some-federal-judges-are-embracing-gen-ai-though-many-are-split-on-its-potential-for-courts/ ↩

  28. *Detroit News* (AP), “Judges are increasingly using AI to draft rulings and prepare for hearings,” April 2, 2026, https://www.detroitnews.com/story/news/nation/2026/04/02/judges-increasingly-using-ai-draft-rulings-prepare-hearings/89433009007/ ↩

  29. Phys.org, “AI tools are widely used by federal judges, study finds,” March 2026, https://phys.org/news/2026-03-ai-tools-widely-federal.html ↩

  30. Federal Rules of Civil Procedure, Rule 11, https://www.law.cornell.edu/rules/frcp/rule_11 ↩

  31. Damien Charlotin, AI Hallucination Cases Database, https://www.damiencharlotin.com/hallucinations/ ↩

  32. Thomson Reuters Institute, “Responsible AI Use for Courts,” January 2026, https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2026/01/Hallucinations-Report-2026_FINAL.pdf ↩

  33. *Couvrette v. Wisnovsky*, No. 3:24-cv-01444-SI (D. Or. Feb. 27, 2026); NWSidebar, https://nwsidebar.wsba.org/2026/03/02/parade-of-horribles-federal-court-in-oregon-surveys-sanctions-for-ai-fake-citations/

    —

    **Continue to [Part 5: The Labeling Problem](./ai-sanctions-wave-part5.md)**

    **Back to [Part 3: The Sixth Circuit Line](./ai-sanctions-wave-part3.md)**

    **Series Index: [AI Sanctions Wave](./ai-sanctions-wave-index.md)** ↩